pip install -e thirdparty/mast3r ERROR

[Triceeee]

hello! how can i fix that
using ubantu 24.04, VMware.
I'm beginner in Linux, thanks for your help.

[teiseong]

I faced the same problem.
This is how I fixed: (Ubuntu 22.04, CUDA 12.4)
Reinstall PyTorch for CUDA 12.4

conda activate mast3r-slam

Clean uninstall of current PyTorch

  pip uninstall torch torchvision torchaudio -y
  conda uninstall pytorch torchvision torchaudio pytorch-cuda -y

Reinstall PyTorch with CUDA 12.4 support

conda install pytorch==2.5.1 torchvision==0.20.1 torchaudio==2.5.1 pytorch-cuda=12.4 -c pytorch -c nvidia

Verify PyTorch installation

python -c "import torch; print(torch.__version__); print(torch.cuda.is_available())"

If verification passes, proceed with mast3r installation

[egirgin]

Here is what worked for me:

pip install torch==2.5.1 torchvision==0.20.1 torchaudio==2.5.1 --index-url https://download.pytorch.org/whl/cu124

conda install \
-c pytorch -c nvidia \
pytorch-cuda=12.4 \
cuda-toolkit=12.4 \
cuda-cccl_linux-64=12.4


cd ./MASt3R-SLAM/thirdparty/mast3r/dust3r/croco/models/curope/
python setup.py build_ext --inplace

[sober-Matthew]

I encountered the same problem. May I ask if you have solved it?

[egirgin]

I encountered the same problem. May I ask if you have solved it?

The problem I had was related to some sort of CUDA directory problem. Since croco has its own layer implementations, it requires cuda-toolkit to run safe and sound. I use a shared computer so there was a problem with the cuda default directory but the code snippet I showed resolved it without manipulating the shared directories.

Given a conda environment, the key is to install pytorch via pip, and install cuda-toolkit (and others) using conda. Then croco installation was able use cuda (or cudnn, not sure) libraries.

Quote of my solution again:

Here is what worked for me:

pip install torch==2.5.1 torchvision==0.20.1 torchaudio==2.5.1 --index-url https://download.pytorch.org/whl/cu124

conda install \
-c pytorch -c nvidia \
pytorch-cuda=12.4 \
cuda-toolkit=12.4 \
cuda-cccl_linux-64=12.4


cd ./MASt3R-SLAM/thirdparty/mast3r/dust3r/croco/models/curope/
python setup.py build_ext --inplace

[sober-Matthew]

As shown below, I still have problems.
python setup.py build_ext --inplace
Traceback (most recent call last):
File "/data02/zhangwei/ysk/MASt3R-SLAM/thirdparty/mast3r/dust3r/croco/models/curope/setup.py", line 5, in
from torch import cuda
File "/home/zw/anaconda3/envs/mast3r-slam/lib/python3.11/site-packages/torch/init.py", line 367, in
from torch._C import * # noqa: F403
^^^^^^^^^^^^^^^^^^^^^^
ImportError: /home/zw/anaconda3/envs/mast3r-slam/lib/python3.11/site-packages/torch/lib/libtorch_cpu.so: undefined symbol: iJIT_NotifyEvent

[wangcong03-shizhuang]

This works for me, by the way, it need pip install -e thirdparty/mast3r --no-build-isolation to install it.

Here is what worked for me:

pip install torch==2.5.1 torchvision==0.20.1 torchaudio==2.5.1 --index-url https://download.pytorch.org/whl/cu124

conda install \
-c pytorch -c nvidia \
pytorch-cuda=12.4 \
cuda-toolkit=12.4 \
cuda-cccl_linux-64=12.4


cd ./MASt3R-SLAM/thirdparty/mast3r/dust3r/croco/models/curope/
python setup.py build_ext --inplace